The Thalamus: The Relay Station Every Sense Except Smell Must Pass

Introduction

There is a structure in the middle of your head, roughly the size and shape of two walnuts pressed together, and almost nothing reaches your conscious mind without its permission.

Look at anything. The light hitting your retina does not go to your visual cortex. It goes to a nucleus in the thalamus first, and only what that nucleus passes on becomes something you see. The same is true of every sound you have heard, every texture you have felt, every flavour, every sense of where your limbs are in the dark. All of it stops in the thalamus before it becomes experience.

Except smell. Odour goes somewhere else, and it gets there first.

That single exception is why the thalamus shows up in almost every popular account of the brain and why almost none of those accounts explain it. Search for the thalamus and you will find a great deal about where it sits, a nucleus-by-nucleus table written for a medical exam, and one sentence about smell being the odd one out. The sentence is always stated and never unpacked.

It deserves unpacking, because the real answer is better than the myth. Smell does not avoid the thalamus. It avoids the thalamic front door. There is a thalamic olfactory route, it runs through a nucleus called the mediodorsal, and human imaging shows traffic on it every time you actually pay attention to a smell [12]. What olfaction skips is the obligatory first stop, not the structure.

And while we are correcting things, the phrase in the title needs correcting too. The thalamus is not a relay. Murray Sherman, who has spent a career on this, eventually published a paper called "The thalamus is more than just a relay" [2], and the wiring backs him up in a way that is hard to argue with. In the visual part of the thalamus, the input from your eyes is a small minority of the synapses. Most of what arrives comes backwards, from cortex. A relay does not get ten times more instruction than it gets input.

So this article does three things. It explains what actually happens when a signal passes through, which is more like a checkpoint than a wire. It takes the smell exception seriously and follows it to the more interesting place it leads. And it shows what the gate does in ordinary life, including the reason some people sleep through a passing siren and others do not, which turns out to be measurable.

Some of this is settled and stated plainly. Some of it is actively argued, and where it is, the argument is described rather than resolved.

Where It Sits and What It Is

The thalamus sits at the top of the brainstem, buried under the cortex, at the geometric centre of the brain. There are two of them, one in each hemisphere, sitting either side of the third ventricle. In many people the two halves are joined across the gap by a bridge of grey matter called the massa intermedia, and in a substantial minority they are not joined at all, which causes no obvious problem.

Each thalamus is not one thing. It is a cluster of dozens of distinct nuclei, separated into groups by a Y-shaped sheet of white matter called the internal medullary lamina. Anterior group in front, medial group inside the fork, lateral group outside it, and a set of specialised nuclei tucked at the back and along the edges.

This matters more than it sounds. Every one of those nuclei has its own inputs, its own cortical target and its own job. A lesion two millimetres across in one nucleus produces a completely different deficit from a lesion two millimetres across in its neighbour. The thalamus behaves less like an organ and more like a switchboard room where each cabinet handles a different line.

The definitive anatomical reference for the human version is the stereotactic atlas built by Anne Morel, Michel Magnin and Daniel Jeanmonod, which mapped the nuclei against multiple staining architectures rather than one [3]. In 2003 Timothy Behrens and colleagues did something that had not been possible before, using diffusion imaging to map thalamic nuclei in living HUMAN brains by tracing which piece of cortex each part connected to [4]. The connection patterns they recovered non-invasively matched the histological picture well enough to use clinically.

More recently, Kai Hwang and colleagues asked a network question rather than an anatomical one: if you treat the whole brain as a set of interacting networks, where does the thalamus sit [5]? Their answer, from HUMAN resting-state imaging, was that it behaves as an integrative hub. It does not belong to one network. It connects across many of them at once, which is exactly the profile you would expect from something that is not merely relaying.

If you want a picture of how the thalamus fits into the wider architecture of a single signal, the mechanics of how neurons communicate at the level of the individual synapse is where the story starts.

The Front Door

Take vision, because it is the best studied case.

Light hits the retina. Retinal ganglion cells send their axons down the optic nerve, through the optic chiasm, and into the lateral geniculate nucleus of the thalamus. The lateral geniculate is a layered structure, six layers in primates, sorted by which eye the input came from and what kind of ganglion cell sent it. Cells in the lateral geniculate then project to primary visual cortex.

Nothing about that route is optional. Destroy the lateral geniculate and the visual cortex receives nothing, regardless of how healthy the eye is.

Hearing works the same way through a different nucleus. Sound is transduced in the cochlea, climbs the brainstem auditory pathway, reaches the inferior colliculus in the midbrain, and then enters the medial geniculate nucleus of the thalamus before continuing to primary auditory cortex.

Touch splits by territory. Sensation from the body, along with pain and temperature, arrives at the ventral posterolateral nucleus. Sensation from the face arrives at the ventral posteromedial nucleus, which also has a small subdivision handling taste. Both send on to primary somatosensory cortex.

Movement uses the same architecture in reverse. Output from the cerebellum and from the basal ganglia does not go straight to motor cortex. It goes through the ventral anterior and ventral lateral nuclei of the thalamus first. When a practised skill runs smoothly, the loop between the cerebellum and procedural memory is passing through thalamic territory on its way to the cortex that executes the movement.

The consistent pattern is this: a sensory or motor signal on its way to cortex enters a specific thalamic nucleus, synapses onto a relay neuron, and continues. The nucleus is dedicated. The route is fixed. And the whole system evolved alongside the six-layer neocortex it feeds.

Which is precisely why smell is different.

The Sense That Skips the Front Door

Odour molecules land on olfactory receptor neurons in a patch of tissue high in the nasal cavity. Those neurons are unusual in that their axons pass directly through the cribriform plate of the skull into the olfactory bulb, which sits underneath the frontal lobe. This is the shortest route from the outside world to the brain that exists anywhere in the body.

From the bulb, mitral and tufted cells send axons down the lateral olfactory tract into primary olfactory cortex, which is a collection of structures rather than a single area: piriform cortex, the anterior olfactory nucleus, the olfactory tubercle, part of the amygdala, and the lateral entorhinal cortex.

At no point in that chain is there a mandatory thalamic stop.

The reason is evolutionary and it is written into the tissue. Piriform cortex is paleocortex, an older three-layer design. The neocortex that vision, hearing and touch report to has six layers and is a much later development. The thalamus, in the form we are describing, evolved in step with that six-layer cortex, as the structure that feeds it. Olfaction had already solved the problem of getting information into a cortex before that arrangement existed, and it kept the solution it had.

Two consequences follow, and both are worth sitting with.

The first is that odour reaches the amygdala and the entorhinal cortex before any thalamic processing has happened. The amygdala is where a signal gets tagged for emotional significance, which is a large part of why smell feels so directly wired to feeling. If you want the tagging mechanism itself, the way the amygdala marks a memory for permanent storage is the same machinery operating on a shorter cable.

The second is a puzzle rather than a fact. Gordon Shepherd laid it out in 2005 in a paper called, plainly enough, "Perception without a Thalamus" [6]. If a thalamic relay is required to generate conscious perception in every other modality, and olfaction largely lacks one, then where does conscious smell perception come from? Shepherd did not claim to have the answer. He argued the question was being ignored rather than solved.

The Olfactory Thalamus Nobody Mentions

Here is where the popular version stops and the interesting part begins.

Piriform cortex projects to the mediodorsal nucleus of the thalamus. The mediodorsal nucleus projects on to orbitofrontal cortex. That is a thalamic olfactory pathway. It is well documented, it is not controversial, and it is missing from essentially every consumer explainer on the subject.

Wendy Tham, Richard Stevenson and Laurie Miller reviewed what the mediodorsal nucleus actually does in olfaction and found it implicated in odour discrimination, odour learning and olfactory attention, with supporting evidence from patients who had thalamic damage [8]. Emmanuelle Courtiol and Donald Wilson went further and recorded from it directly in RAT, showing that a subset of mediodorsal neurons respond selectively to odours [10], and later that mediodorsal activity tracks odour-guided behaviour rather than just odour presence [11].

The HUMAN evidence is the piece that settles the argument for a general reader. Jane Plailly, working with Jay Gottfried's group, used fMRI to test what happens when people simply direct their attention toward a smell rather than toward something else [12]. Attending to an odour increased the effective connectivity between mediodorsal thalamus and both piriform and orbitofrontal cortex. This was a small single-site imaging study, of the size normal for olfactory fMRI at the time, so the finding should be held as a demonstration rather than a population estimate. But the direction is clear. A structure with no role in a sense does not change its coupling when you attend to that sense.

Then there is the most radical reading, and it comes from the people who understand thalamic relays best. Leslie Kay and Murray Sherman published a paper in 2007 titled "An argument for an olfactory thalamus" [7]. Their case is that the olfactory bulb itself performs the computations a thalamic relay performs. Same job, different structure, arrived at independently. On that account olfaction does not lack a thalamus. It has one that nobody labelled as such.

This is a reframing, not a consensus, and it should be attributed to Kay and Sherman by name rather than presented as established. What is genuinely open is narrower and sharper: how necessary the mediodorsal route is for conscious smell. Courtiol and Wilson titled their 2015 review "unanswered questions about the role of the mediodorsal thalamic nucleus in olfaction" and meant it [9].

The most direct HUMAN test comes from people who have thalamic damage. Lee Sela, working with Noam Sobel's group, tested olfactory ability in patients with thalamic lesions and found a split rather than a blanket loss [62]. Basic detection was largely spared. What suffered were the judgements that require the odour to be evaluated rather than merely registered, including hedonic judgement and olfactory attention.

That is the shape of the answer, as far as anyone has one. You do not need the thalamus to notice that something smells. You appear to need it to work out what you think about the smell. Whether that pattern survives further testing is not decided, and the honest summary is that the necessity question is still open.

Some perspective on how much of smell runs below awareness anyway: Noam Sobel's group has argued that HUMAN olfaction operates in something close to a constant state of change blindness, with most odour information never reaching report at all [14]. And the wider mapping of how odour reaches orbitofrontal cortex in humans, reviewed by Jay Gottfried and David Zald, shows the picture was never as simple as one direct line [13].

So the defensible sentence, the one this article stands behind, is this. Olfaction is the only sense without an obligatory pre-cortical thalamic relay. It is not the only sense the thalamus ignores.

The Relay That Barely Listens to Its Own Input

Now the second correction, and it is the one that changes how you should think about the whole structure.

Go back to the lateral geniculate nucleus, the visual relay. If it were genuinely a relay, you would expect the retina to be its dominant input. Count the synapses and that is not what you find. Retinal terminals make up a small minority of the synapses on lateral geniculate relay cells. The rest come from three other places: feedback projections descending from layer 6 of visual cortex, inhibitory input from the thalamic reticular nucleus and local interneurons, and modulatory input from the brainstem.

The synapse census work was done carefully in CAT by Susan Van Horn, Alev Erisir and Murray Sherman using quantitative electron microscopy [15], building on Sherman and Christof Koch's earlier analysis of what actually controls transmission from retina to cortex [16]. The figure that keeps coming back is that retinal terminals account for something in the region of five to ten percent of the synapses on relay cells, with descending cortical feedback supplying the largest single share of everything else. Fibres running backwards from cortex to thalamus outnumber those running forwards by roughly an order of magnitude.

Treat both of those as approximations. They vary by species, by pathway and by counting method, and anyone who quotes them to a decimal place is inventing precision that the anatomy does not support. The order of magnitude is the point, and the order of magnitude is not in doubt.

What is the feedback for? Farran Briggs and Martin Usrey reviewed the corticothalamic system and concluded it modulates gain, timing and spatial tuning of what passes through rather than carrying content itself [17]. Javier Cudeiro and Adam Sillito, looking back over the visual case, described the same thing as cortex actively shaping the input it is about to receive [18]. Cortex is not waiting to be told what is out there. It is leaning on the gate.

The conceptual tool that makes all of this tractable came from Sherman and Ray Guillery in 1998, in a paper about distinguishing drivers from modulators [1]. Some inputs to a neuron carry the message. Others only adjust how the message is handled. They differ in where they synapse on the cell, in what receptors they use, and in how they behave when stimulated repeatedly. Once you separate the two, the picture resolves: the retina is a driver, a minority input in number but the one carrying content, and almost everything else arriving at the lateral geniculate is a modulator with a vote on whether the content gets through.

Sherman and Guillery developed the argument across two further papers on how information actually flows to cortex [19] [20], and Sherman eventually made the strongest version of the claim in 2016: the thalamus is not peripheral to cortical function but central to it, participating continuously rather than at the input stage [21]. Michael Halassa and Sherman later set out a general framework for the recurring circuit motifs involved [22].

The driver and modulator distinction also produces a second, very useful split.

First Order and Higher Order

Some thalamic nuclei get their driving input from the periphery. The lateral geniculate from the retina, the medial geniculate from the auditory brainstem, the ventral posterior nuclei from the ascending touch pathways. These are called first order relays and they do the job the textbook describes.

Other nuclei get their driving input from cortex itself. The pulvinar, the mediodorsal nucleus and the posterior medial nucleus all receive their content-carrying input from one cortical area and pass it to another. These are higher order relays, and they mean something specific: when one part of cortex talks to another part of cortex, one of the routes it uses runs down into the thalamus and back up again.

That is a strange design if you think of the thalamus as an input stage. It makes complete sense if you think of it as a switchboard that cortex uses to route its own internal traffic.

Higher order relays also behave differently. Ramcharan, Gnadt and Sherman found in MACAQUE that higher order relays fire in burst mode considerably more often than first order relays do [25], which brings us to the other thing thalamic neurons do that a wire cannot.

Two Ways to Fire

A thalamic relay neuron has two firing modes, and which one it is in changes what gets through.

In tonic mode the cell fires in a roughly linear way. Stronger input produces more spikes, weaker input produces fewer, and the relationship is faithful enough that cortex receives something close to a fair copy of what arrived. This is the mode of alert wakefulness.

In burst mode the cell has been hyperpolarised for long enough to de-inactivate a particular calcium channel, and a subsequent input triggers not a proportional response but a short high-frequency volley of spikes. Burst mode is a terrible reporter and an excellent alarm. It is nonlinear, so it loses the fine detail of the input, but it is extremely good at getting cortex to notice that something happened at all.

Sherman laid out the two modes and their consequences in 2001 [23]. Mircea Steriade and Rodolfo Llinás had already established the broader picture in 1988: the thalamus has functional states, those states track the sleep and wake cycle, and the neuronal interplay that produces them is intrinsic to thalamic circuitry rather than imposed from outside [24].

This is where the popular claim that the thalamus shuts down during sleep goes wrong. It does not shut down. It switches mode. The switch is active, it is generated locally, and it has a specific circuit behind it.

The Gate Has a Name

Wrapped around the outside of the thalamus, like a thin shell, is a sheet of neurons called the thalamic reticular nucleus. Every fibre running from thalamus to cortex passes through it. Every fibre running from cortex back down to thalamus passes through it too. Each one gives off a branch as it goes.

The reticular nucleus is entirely inhibitory. And it has one further property that decides its function: it sends nothing at all to cortex. Its output goes back into the thalamus.

So here is a structure that receives a copy of everything moving in both directions, and whose only power is to suppress thalamic transmission. That is not a relay component. That is a gate.

Francis Crick worked out the implication in 1984, before most of the supporting evidence existed, and proposed that the reticular nucleus acts as an attentional searchlight, selectively releasing some thalamic channels from inhibition while holding others shut [26]. Didier Pinault's later review of its structure and function set out how much of Crick's framing held up and where it needed refining [27].

The modern work is more specific. Michael Halassa and colleagues showed in MOUSE that the reticular nucleus is not one uniform sheet but a set of subnetworks whose architecture changes with behavioural state, with different populations engaged during sleep and during attention [28]. Basilis Zikopoulos and Helen Barbas traced, in MACAQUE, a projection from prefrontal cortex to the reticular nucleus with a pattern unlike any other cortical input to it, which is exactly the anatomy an executive attention system would need if it wanted to reach down and adjust the gate directly [29]. Laura Lewis and colleagues showed the effect can be strikingly local: stimulating a small part of the reticular nucleus in MOUSE produced fast, spatially restricted changes in arousal state rather than a whole-brain switch [30].

That last result is worth pausing on, because it means the gate is not a single lever. Parts of your brain can be more asleep than others.

Why You Can Sleep Through Traffic

Fall asleep and your electroencephalogram starts producing spindles: short bursts of oscillation at roughly eleven to sixteen hertz, lasting from about half a second up to a few seconds, characteristic of the second stage of non-REM sleep.

Spindles are made in the thalamus. The reticular nucleus and the thalamocortical relay cells form a loop, the reticular cells inhibit the relay cells, the relay cells rebound with a burst, and the interaction between the two produces the rhythm. Steriade, David McCormick and Terrence Sejnowski described the whole family of thalamocortical oscillations and their relation to sleep and arousal in 1993 [31], and the modern account of spindle mechanism and function was reviewed at length by Laura Fernandez and Anita Lüthi [32] [33].

Now the finding that makes this concrete.

Thien Thanh Dang-Vu and colleagues had people sleep in a laboratory while sounds were played to them, and asked which sleepers kept sleeping [34]. The answer tracked spindles. People whose brains produced spindles at a higher rate were more likely to sleep through the noise. People with fewer spindles woke up. This was a small laboratory study, so it should be read as a demonstration of the mechanism rather than as a population statistic, and no participant count is quoted here because the figure could not be independently confirmed for this article.

But the direction of the result is the thing. The reason one person sleeps through a siren and their partner does not is not a personality trait or a matter of willpower. It is a property of a thalamic circuit, and it is measurable on an electrode.

The gate also has a positive job during the night, not just a defensive one. Spindles coincide with the reactivation of newly learned material, which is why they turn up repeatedly in the literature on how sleep consolidates spaced learning. The same circuitry that shuts the door on the outside world is busy on the inside.

And the control runs in both directions. Thomas Gent, Mojtaba Bandarabadi, Carolina Gutierrez Herrera and Antoine Adamantidis showed in MOUSE that the centromedial thalamus can drive the transition into sleep or into wakefulness depending on how it is stimulated [35]. Slow stimulation put animals to sleep. Fast stimulation woke them up. The same nucleus, opposite outcomes.

Attention Happens Before Cortex

For a long time attention was treated as something cortex did. You looked at a scene, cortex decided what mattered, and the decision was made high in the processing chain.

The thalamic work makes that story hard to keep.

Kerry McAlonan, James Cavanaugh and Robert Wurtz recorded from both the lateral geniculate nucleus and the reticular nucleus in MACAQUE while the animals directed attention to one location or another [36]. Both structures showed attentional modulation. The ordering is what mattered: the effect appeared in the reticular nucleus before it appeared in the geniculate. That is the sequence a gating account predicts and the opposite of what you would expect if the thalamus were simply inheriting a decision made upstairs.

In HUMAN subjects, Daniel O'Connor, Miki Fukui, Mark Pinsk and Sabine Kastner had already shown with fMRI that attention modulates responses in the lateral geniculate nucleus itself [37]. Attending to a stimulus changed the signal in the visual relay, not just in visual cortex.

Ralf Wimmer and colleagues then went after causation in MOUSE, showing that the reticular nucleus is required for sensory selection when attention has to be divided between competing streams [38]. Interfering with it did not blunt perception in general. It broke the ability to choose.

Ian Schmitt, Ralf Wimmer, Miho Nakajima and Halassa found a further role for the mediodorsal nucleus, showing in MOUSE that it amplifies connectivity between cortical neurons in a way that sustains attentional control over time [39]. Not carrying the content of the attended signal. Holding the cortical configuration that keeps attending possible.

Halassa and Kastner drew the strands together into an argument that thalamic contributions to cognitive control are distributed and structural rather than incidental [40], and Miho Nakajima and Halassa reviewed the evidence that thalamus sets functional connectivity in cortex rather than merely feeding it [41].

The practical upshot is that focus is not purely a cortical achievement. Anyone interested in the relationship between attention and memory is looking at a system whose first filter sits below cortex, and which has already discarded most of the input before the question of what to remember arises. What survives the gate is roughly what sensory memory holds for a fraction of a second before most of that goes too.

The Pulvinar: Cortex Talking to Cortex Through the Middle

The pulvinar is the largest nucleus in the primate thalamus, and it has no peripheral input at all. It talks almost exclusively to cortex. It also expanded dramatically over primate evolution, which is not what you would expect of a structure whose job was passing sensation along.

Yuri Saalmann, Mark Pinsk, Liang Wang, Xin Li and Kastner tested what it does in MACAQUE by recording simultaneously from the pulvinar and from two visual cortical areas while attention shifted [42]. When attention was directed to a location, the pulvinar synchronised activity between the two cortical areas in a way that tracked the attentional demand. It was regulating the transmission of information between cortical regions rather than supplying information to them.

Huihui Zhou, Robert Schafer and Robert Desimone extended the causal case, showing that pulvinar activity is necessary for normal attentional modulation in visual cortex rather than merely correlated with it [43].

Put next to the higher order relay concept, the pulvinar becomes the clearest single example of the thalamus doing something a relay cannot do. It is a routing layer for cortex, sitting underneath cortex, deciding which cortical conversations get bandwidth.

There is a structural reason the thalamus can do this at all, and it comes from Edward Jones. He described two classes of thalamocortical cell distinguished by their calcium-binding protein and their projection pattern [44]. Core cells, rich in parvalbumin, project precisely to layer 4 of one cortical column and carry specific content. Matrix cells, rich in calbindin, project diffusely to superficial layers across wide areas of cortex. The core system delivers the message. The matrix system can synchronise large stretches of cortex at once. A structure with both can be a precise relay and a global coordinator using different cells in the same nucleus.

A Map of the Nuclei, by Job

Anatomy pages give you the nuclei as a list of names and positions. It is more useful to read them as a list of jobs.

The last row is the one to remember. Every other nucleus in that table sends something to cortex. The reticular nucleus does not. It exists solely to decide what the others are allowed to send.

Memory, Thought and the Thalamus

The memory nuclei deserve their own treatment, because thalamic amnesia is real and is routinely mistaken for hippocampal amnesia.

Stuart Zola-Morgan and Larry Squire showed in MONKEY that lesions of the mediodorsal nucleus produce memory impairment [48]. This was a small primate lesion study of its period, so no animal count is quoted here, but the finding has been repeatedly supported since. John Aggleton and colleagues asked a harder question about the anterior thalamic nuclei: why do lesions there cause spatial memory deficits as severe as they do, given how small the structures are [47]? Their answer is that the anterior nuclei are not a waystation on a memory circuit but a hub within it, so removing them degrades the whole network rather than cutting one line.

Anna Mitchell reviewed the mediodorsal nucleus specifically as a higher order relay serving learning and decision-making [45], and Mathieu Wolff and Seralynne Vann made the broader case for what they called the cognitive thalamus: a set of nuclei that support mental representation rather than sensory transmission [46].

The nucleus reuniens sits on the midline and connects medial prefrontal cortex with the hippocampus, described in detail by Robert Vertes and colleagues [49]. It is the anatomical bridge between the structure that plans and the structure that remembers, which is a large part of why thalamic damage can produce deficits that look simultaneously frontal and amnesic.

The most striking recent result is causal. Zengcai Guo, Hidehiko Inagaki, Karel Svoboda and colleagues found in MOUSE that when frontal cortex holds a plan in mind across a delay, the persistent activity that constitutes the holding does not live in cortex alone [50]. Silence the thalamus and the persistent activity collapses. The loop between thalamus and cortex is the thing that maintains it, not the cortex by itself.

That reframes short-term memory in a way worth stating plainly. Holding a thought is not something cortex does while the thalamus waits. It is a circuit running through both. The relationship between the prefrontal cortex and working memory is a thalamocortical relationship, and the mediodorsal nucleus is the prefrontal cortex's dedicated partner in it. Likewise, the route by which the hippocampus decides what to remember passes through the anterior thalamic nuclei on its way around the Papez circuit.

The Consciousness Question

Now the part that has to be handled carefully, because it is genuinely important and genuinely unsettled.

In 2007 Nicholas Schiff and colleagues published a case in Nature that changed the conversation [51]. A man in his late thirties had been in a minimally conscious state for six years following a severe traumatic brain injury. They implanted electrodes for deep brain stimulation in his central thalamus. With stimulation, he showed improvements in arousal, in limb control, in oral feeding and in consistent communication.

The sample size was one patient. That number has to be stated every time this result is cited, because a single case presented without its denominator is how a genuinely important finding becomes an overclaim. What the case established was that stimulating the central thalamus can change conscious behaviour in a person, which is a different and more limited claim than saying consciousness lives there. Schiff has since written a fuller account of the rationale and its limits [52].

Animal work has since given the mechanism more shape. Michelle Redinbaugh and colleagues showed in MACAQUE that the central lateral thalamus modulates conscious state through layer-specific control of cortex, with stimulation of deep cortical layers restoring wakeful-type activity [53]. André Bastos and colleagues did the complementary experiment, anaesthetising MACAQUE with propofol and then reversing the unconsciousness with thalamic stimulation while recording the cortical dynamics that came back with it [54]. This was a small number of animals, as primate electrophysiology always is.

Rodolfo Llinás and colleagues approached it from the pathological direction, proposing thalamocortical dysrhythmia as a common mechanism behind a family of neurological and psychiatric syndromes in which the thalamocortical loop settles into the wrong rhythm [55].

Where does that leave the claim? The honest position is that thalamocortical loops are a leading candidate substrate for conscious awareness, and that the causal evidence for the central thalamus being able to switch conscious states is now strong. That is not the same as showing consciousness resides there. Competing accounts put considerably more weight on cortex, and the field has not settled it. Anyone who tells you the thalamus is the seat of consciousness is ahead of the evidence.

What Breaks

Thalamic strokes are small and their effects are disproportionate. Jeremy Schmahmann's review organised them by arterial territory, which is the only way the clinical picture makes sense [56].

Emmanuel Carrera and Julien Bogousslavsky examined how anatomically distinct thalamic strokes produce distinct behavioural syndromes, which reinforces the point that the thalamus is many structures wearing one name [57].

The inferolateral territory produces the strangest consequence. Damage to the sensory relay can leave the affected region numb and then, weeks or months later, intensely and persistently painful. This is central post-stroke pain, historically called Dejerine-Roussy syndrome, reviewed by Henriette Klit, Nanna Finnerup and Troels Jensen [58]. The pain is generated by damage to the pain pathway itself rather than by any injury in the body part that hurts. Prevalence figures for it vary widely depending on cohort and case definition, so none is quoted here.

Language can break too. Thalamic aphasia is well described, though its mechanism is argued over. Bruce Crosson reconsidered the thalamic contribution to language in light of newer findings and concluded the thalamus participates in selecting and maintaining lexical-semantic material rather than housing language itself [59].

When the Sleep Gate Is Destroyed

There is one disease that demonstrates the thalamic role in sleep more completely than any experiment could, and it should be described soberly.

In 1986 Elio Lugaresi and colleagues reported a familial condition in the New England Journal of Medicine in which patients lost the ability to sleep, developed autonomic disturbance, and died [60]. On post-mortem examination the degeneration was not spread across the brain. It was concentrated in the thalamus, specifically in the anterior ventral and mediodorsal nuclei. The condition was named fatal familial insomnia.

It was later established as a prion disease, and Pasquale Montagna and colleagues reviewed both the familial and the sporadic forms [61]. It is extremely rare, it is genetic in the familial form, and it has no treatment.

Two things need saying alongside it. The first is that it has nothing to do with ordinary insomnia. Difficulty sleeping is common and its causes are almost entirely mundane. The second is what the disease demonstrates scientifically: when specific thalamic nuclei are destroyed, sleep does not become difficult, it becomes impossible. The circuit that generates spindles and gates sensory input during the night is not a contributor to sleep. It is a requirement for it.

What Is Settled and What Is Not

Enough of the above is contested that it is worth separating clearly.

Conclusion

The sentence in the title is the one everybody knows, and both halves of it turn out to be worth arguing with.

"Relay station" undersells it badly. The structure that vision passes through gets most of its synapses from somewhere other than the eye, spends most of its wiring listening to cortex, has two firing modes with different consequences, and sits inside an inhibitory shell that can close individual channels while leaving others open. Nothing about that is passive. Sherman's title from 2007 was the right summary, and everything found since has made it more so.

"Every sense except smell" is closer to true, and still not the whole story. Olfaction does skip the obligatory first relay, and the reason goes back to a cortex older than the one the thalamus evolved to feed. But odour information reaches the thalamus later, through the mediodorsal nucleus, and the traffic on that route goes up when you actually pay attention to what you are smelling. Whether the route is necessary for conscious smell is still an open question, and it is a better question than the tidy version it replaced.

What is left, once both corrections are made, is a structure doing something more interesting than either phrase suggests. It decides what you get to know about. It does that continuously, in conversation with the cortex it reports to, using a gate that has a name and a mechanism. When it works, you never notice it, which is why almost nobody has heard of the reticular nucleus. When it fails, the failure is not subtle: pain with no injury, language that will not assemble, or a sleep that never comes.

Two walnuts in the middle of your head, holding the door.